Well, as a matter of fact, Tandem's NonStop Kernel (yes, it's
still called that) provides both software and hardware fault
tolerance, as well as near-linear scalability. It is an
entirely different approach than hardware fault tolerance,
though it includes hardware fault tolerance too where that
makes sense.
As to more reliable than MVS, yes, it's much more reliable
exactly because of the software fault tolerance, which includes
tolerance of OS faults and processor halts. That's not to
say than any given instance of NSK is more reliable than
an instance of MVS, but that systems running NSK are far more
reliable than clusters running MVS (BTW, all systems running NSK
are clusters). NSK systems survive any single point of failure,
hardware or software, processor or
peripheral, with data integrity and user transparancy.
But the real point for most business applications isn't the
extreme level of fault tolerance. It's scalability
(as long as you have data integrity, which MVS and anybody else
can achieve without the exreme NSK level of fault tolerance).
These massively parallel systems can roll up more computing
power than any mainframe cluster exactly because they scale
in a near linear way owing to their loosely coupled
architecture.
Nothing prevents anyone else from developing a loosely coupled
cluster, but there are extreme difficulties in doing so, ALWAYS greatly underestimated by them what ain't tried it.
Tandem's been doing it for over 20 years. Cpq will capitalize
on that. |
